Article: Google signed a power-purchase agreement with geothermal developer Fervo for up to 400 MW of electricity from a Utah project slated to start delivering energy in 2028. The deal gives the tech giant a steady, carbon-free source of electricity for its expanding AI workloads and nudges the cloud-computing sector toward baseload renewables.

Why baseload power matters for AI

Training large language models and running inference at scale guzzles megawatts of electricity for hours on end. Unlike web traffic that ebbs and flows, AI training jobs run nonstop until a model converges. Wind farms and solar arrays can’t promise that kind of constant output; they swing with weather and time of day.

Geothermal energy, by contrast, spits out heat-derived electricity 24 hours a day, every day. Buying 400 MW of geothermal power gives Google a reliable “baseload” supply that matches the uninterrupted demand of its GPU farms, cutting the need for massive battery banks.

How enhanced geothermal fits the bill

Fervo’s Utah plant uses an enhanced geothermal system (EGS). Traditional geothermal sites sit on naturally porous, hot rock that lets water circulate to the surface. EGS drills deep into hot, impermeable rock, injects water at high pressure, and cracks the rock to create artificial pathways for heat exchange. The trick works where conventional geothermal resources are absent, widening the geographic options for clean baseload power.

The closed-loop design recirculates water through the fracture network, extracts heat each pass, then sends the fluid back underground. Because the water never mixes with the surrounding rock, groundwater stays protected and water consumption stays low.

Financial ripple

Fervo’s shares jumped 30 % after the announcement, signalling investor confidence in the commercial viability of EGS. Drilling deep wells costs a lot upfront, but once the plant is online, operating expenses are modest compared with fossil-fuel plants.

Stakes for the broader cloud market

If the Utah project proves reliable, rivals such as Microsoft and Amazon may chase similar EGS deals. A shift toward geothermal would diversify the clean-energy mix feeding AI supercomputers and could speed the tech sector’s overall decarbonisation.

Challenges and what to watch

Key indicators to follow include:

  • Construction milestones – hitting the 2028 commissioning date on schedule will validate the business case.
  • Power-delivery data – actual capacity factor and availability will show whether geothermal can truly replace intermittent sources for AI loads.
  • Industry response – announcements from other cloud providers about geothermal purchases will reveal whether Google’s move sparks a broader trend.
  • Policy shifts – new incentives or permitting reforms for geothermal could lower drilling costs and accelerate future projects.

If the Utah plant delivers the promised 400 MW of constant, carbon-free electricity, it will give Google a tangible tool to meet its AI-driven power surge while staying on track for net-zero. The real test will be whether the technology scales beyond a single project and reshapes how the biggest data-center operators think about clean baseload power.